Time-reversible molecular dynamics algorithms with bond constraints are derived. The algorithms are stable with and without a thermostat and in double precision as well as in single-precision arithmetic. Time reversibility is achieved by applying a central-difference expression for the velocities in the expression for Gauss' principle of least constraint. The imposed time symmetry results in a quadratic expression for the Lagrange multiplier. For a system of complex molecules with connected constraints the corresponding set of coupled quadratic equations is easily solved by a consecutive iteration scheme. The algorithms were tested on two models. One is a dumbbell model of Toluene, the other system consists of molecules with four connected constraints forming a triangle and a branch point of constraints. The equilibrium particle distributions and the mean-square particle displacements for the dumbbell model were compared to the corresponding functions obtained by GROMACS. The agreement is perfect within statistical error.